Merge branch 'master' of git://git.kernel.org/pub/scm/linux/kernel/git/jk/spufs into...
[linux-2.6] / fs / dlm / rcom.c
1 /******************************************************************************
2 *******************************************************************************
3 **
4 **  Copyright (C) Sistina Software, Inc.  1997-2003  All rights reserved.
5 **  Copyright (C) 2005-2008 Red Hat, Inc.  All rights reserved.
6 **
7 **  This copyrighted material is made available to anyone wishing to use,
8 **  modify, copy, or redistribute it subject to the terms and conditions
9 **  of the GNU General Public License v.2.
10 **
11 *******************************************************************************
12 ******************************************************************************/
13
14 #include "dlm_internal.h"
15 #include "lockspace.h"
16 #include "member.h"
17 #include "lowcomms.h"
18 #include "midcomms.h"
19 #include "rcom.h"
20 #include "recover.h"
21 #include "dir.h"
22 #include "config.h"
23 #include "memory.h"
24 #include "lock.h"
25 #include "util.h"
26
27
28 static int rcom_response(struct dlm_ls *ls)
29 {
30         return test_bit(LSFL_RCOM_READY, &ls->ls_flags);
31 }
32
33 static int create_rcom(struct dlm_ls *ls, int to_nodeid, int type, int len,
34                        struct dlm_rcom **rc_ret, struct dlm_mhandle **mh_ret)
35 {
36         struct dlm_rcom *rc;
37         struct dlm_mhandle *mh;
38         char *mb;
39         int mb_len = sizeof(struct dlm_rcom) + len;
40
41         mh = dlm_lowcomms_get_buffer(to_nodeid, mb_len, ls->ls_allocation, &mb);
42         if (!mh) {
43                 log_print("create_rcom to %d type %d len %d ENOBUFS",
44                           to_nodeid, type, len);
45                 return -ENOBUFS;
46         }
47         memset(mb, 0, mb_len);
48
49         rc = (struct dlm_rcom *) mb;
50
51         rc->rc_header.h_version = (DLM_HEADER_MAJOR | DLM_HEADER_MINOR);
52         rc->rc_header.h_lockspace = ls->ls_global_id;
53         rc->rc_header.h_nodeid = dlm_our_nodeid();
54         rc->rc_header.h_length = mb_len;
55         rc->rc_header.h_cmd = DLM_RCOM;
56
57         rc->rc_type = type;
58
59         spin_lock(&ls->ls_recover_lock);
60         rc->rc_seq = ls->ls_recover_seq;
61         spin_unlock(&ls->ls_recover_lock);
62
63         *mh_ret = mh;
64         *rc_ret = rc;
65         return 0;
66 }
67
68 static void send_rcom(struct dlm_ls *ls, struct dlm_mhandle *mh,
69                       struct dlm_rcom *rc)
70 {
71         dlm_rcom_out(rc);
72         dlm_lowcomms_commit_buffer(mh);
73 }
74
75 /* When replying to a status request, a node also sends back its
76    configuration values.  The requesting node then checks that the remote
77    node is configured the same way as itself. */
78
79 static void make_config(struct dlm_ls *ls, struct rcom_config *rf)
80 {
81         rf->rf_lvblen = cpu_to_le32(ls->ls_lvblen);
82         rf->rf_lsflags = cpu_to_le32(ls->ls_exflags);
83 }
84
85 static int check_config(struct dlm_ls *ls, struct dlm_rcom *rc, int nodeid)
86 {
87         struct rcom_config *rf = (struct rcom_config *) rc->rc_buf;
88         size_t conf_size = sizeof(struct dlm_rcom) + sizeof(struct rcom_config);
89
90         if ((rc->rc_header.h_version & 0xFFFF0000) != DLM_HEADER_MAJOR) {
91                 log_error(ls, "version mismatch: %x nodeid %d: %x",
92                           DLM_HEADER_MAJOR | DLM_HEADER_MINOR, nodeid,
93                           rc->rc_header.h_version);
94                 return -EPROTO;
95         }
96
97         if (rc->rc_header.h_length < conf_size) {
98                 log_error(ls, "config too short: %d nodeid %d",
99                           rc->rc_header.h_length, nodeid);
100                 return -EPROTO;
101         }
102
103         if (le32_to_cpu(rf->rf_lvblen) != ls->ls_lvblen ||
104             le32_to_cpu(rf->rf_lsflags) != ls->ls_exflags) {
105                 log_error(ls, "config mismatch: %d,%x nodeid %d: %d,%x",
106                           ls->ls_lvblen, ls->ls_exflags, nodeid,
107                           le32_to_cpu(rf->rf_lvblen),
108                           le32_to_cpu(rf->rf_lsflags));
109                 return -EPROTO;
110         }
111         return 0;
112 }
113
114 static void allow_sync_reply(struct dlm_ls *ls, uint64_t *new_seq)
115 {
116         spin_lock(&ls->ls_rcom_spin);
117         *new_seq = ++ls->ls_rcom_seq;
118         set_bit(LSFL_RCOM_WAIT, &ls->ls_flags);
119         spin_unlock(&ls->ls_rcom_spin);
120 }
121
122 static void disallow_sync_reply(struct dlm_ls *ls)
123 {
124         spin_lock(&ls->ls_rcom_spin);
125         clear_bit(LSFL_RCOM_WAIT, &ls->ls_flags);
126         clear_bit(LSFL_RCOM_READY, &ls->ls_flags);
127         spin_unlock(&ls->ls_rcom_spin);
128 }
129
130 int dlm_rcom_status(struct dlm_ls *ls, int nodeid)
131 {
132         struct dlm_rcom *rc;
133         struct dlm_mhandle *mh;
134         int error = 0;
135
136         ls->ls_recover_nodeid = nodeid;
137
138         if (nodeid == dlm_our_nodeid()) {
139                 rc = ls->ls_recover_buf;
140                 rc->rc_result = dlm_recover_status(ls);
141                 goto out;
142         }
143
144         error = create_rcom(ls, nodeid, DLM_RCOM_STATUS, 0, &rc, &mh);
145         if (error)
146                 goto out;
147
148         allow_sync_reply(ls, &rc->rc_id);
149         memset(ls->ls_recover_buf, 0, dlm_config.ci_buffer_size);
150
151         send_rcom(ls, mh, rc);
152
153         error = dlm_wait_function(ls, &rcom_response);
154         disallow_sync_reply(ls);
155         if (error)
156                 goto out;
157
158         rc = ls->ls_recover_buf;
159
160         if (rc->rc_result == -ESRCH) {
161                 /* we pretend the remote lockspace exists with 0 status */
162                 log_debug(ls, "remote node %d not ready", nodeid);
163                 rc->rc_result = 0;
164         } else
165                 error = check_config(ls, rc, nodeid);
166         /* the caller looks at rc_result for the remote recovery status */
167  out:
168         return error;
169 }
170
171 static void receive_rcom_status(struct dlm_ls *ls, struct dlm_rcom *rc_in)
172 {
173         struct dlm_rcom *rc;
174         struct dlm_mhandle *mh;
175         int error, nodeid = rc_in->rc_header.h_nodeid;
176
177         error = create_rcom(ls, nodeid, DLM_RCOM_STATUS_REPLY,
178                             sizeof(struct rcom_config), &rc, &mh);
179         if (error)
180                 return;
181         rc->rc_id = rc_in->rc_id;
182         rc->rc_seq_reply = rc_in->rc_seq;
183         rc->rc_result = dlm_recover_status(ls);
184         make_config(ls, (struct rcom_config *) rc->rc_buf);
185
186         send_rcom(ls, mh, rc);
187 }
188
189 static void receive_sync_reply(struct dlm_ls *ls, struct dlm_rcom *rc_in)
190 {
191         spin_lock(&ls->ls_rcom_spin);
192         if (!test_bit(LSFL_RCOM_WAIT, &ls->ls_flags) ||
193             rc_in->rc_id != ls->ls_rcom_seq) {
194                 log_debug(ls, "reject reply %d from %d seq %llx expect %llx",
195                           rc_in->rc_type, rc_in->rc_header.h_nodeid,
196                           (unsigned long long)rc_in->rc_id,
197                           (unsigned long long)ls->ls_rcom_seq);
198                 goto out;
199         }
200         memcpy(ls->ls_recover_buf, rc_in, rc_in->rc_header.h_length);
201         set_bit(LSFL_RCOM_READY, &ls->ls_flags);
202         clear_bit(LSFL_RCOM_WAIT, &ls->ls_flags);
203         wake_up(&ls->ls_wait_general);
204  out:
205         spin_unlock(&ls->ls_rcom_spin);
206 }
207
208 int dlm_rcom_names(struct dlm_ls *ls, int nodeid, char *last_name, int last_len)
209 {
210         struct dlm_rcom *rc;
211         struct dlm_mhandle *mh;
212         int error = 0;
213         int max_size = dlm_config.ci_buffer_size - sizeof(struct dlm_rcom);
214
215         ls->ls_recover_nodeid = nodeid;
216
217         if (nodeid == dlm_our_nodeid()) {
218                 ls->ls_recover_buf->rc_header.h_length =
219                         dlm_config.ci_buffer_size;
220                 dlm_copy_master_names(ls, last_name, last_len,
221                                       ls->ls_recover_buf->rc_buf,
222                                       max_size, nodeid);
223                 goto out;
224         }
225
226         error = create_rcom(ls, nodeid, DLM_RCOM_NAMES, last_len, &rc, &mh);
227         if (error)
228                 goto out;
229         memcpy(rc->rc_buf, last_name, last_len);
230
231         allow_sync_reply(ls, &rc->rc_id);
232         memset(ls->ls_recover_buf, 0, dlm_config.ci_buffer_size);
233
234         send_rcom(ls, mh, rc);
235
236         error = dlm_wait_function(ls, &rcom_response);
237         disallow_sync_reply(ls);
238  out:
239         return error;
240 }
241
242 static void receive_rcom_names(struct dlm_ls *ls, struct dlm_rcom *rc_in)
243 {
244         struct dlm_rcom *rc;
245         struct dlm_mhandle *mh;
246         int error, inlen, outlen, nodeid;
247
248         nodeid = rc_in->rc_header.h_nodeid;
249         inlen = rc_in->rc_header.h_length - sizeof(struct dlm_rcom);
250         outlen = dlm_config.ci_buffer_size - sizeof(struct dlm_rcom);
251
252         error = create_rcom(ls, nodeid, DLM_RCOM_NAMES_REPLY, outlen, &rc, &mh);
253         if (error)
254                 return;
255         rc->rc_id = rc_in->rc_id;
256         rc->rc_seq_reply = rc_in->rc_seq;
257
258         dlm_copy_master_names(ls, rc_in->rc_buf, inlen, rc->rc_buf, outlen,
259                               nodeid);
260         send_rcom(ls, mh, rc);
261 }
262
263 int dlm_send_rcom_lookup(struct dlm_rsb *r, int dir_nodeid)
264 {
265         struct dlm_rcom *rc;
266         struct dlm_mhandle *mh;
267         struct dlm_ls *ls = r->res_ls;
268         int error;
269
270         error = create_rcom(ls, dir_nodeid, DLM_RCOM_LOOKUP, r->res_length,
271                             &rc, &mh);
272         if (error)
273                 goto out;
274         memcpy(rc->rc_buf, r->res_name, r->res_length);
275         rc->rc_id = (unsigned long) r;
276
277         send_rcom(ls, mh, rc);
278  out:
279         return error;
280 }
281
282 static void receive_rcom_lookup(struct dlm_ls *ls, struct dlm_rcom *rc_in)
283 {
284         struct dlm_rcom *rc;
285         struct dlm_mhandle *mh;
286         int error, ret_nodeid, nodeid = rc_in->rc_header.h_nodeid;
287         int len = rc_in->rc_header.h_length - sizeof(struct dlm_rcom);
288
289         error = create_rcom(ls, nodeid, DLM_RCOM_LOOKUP_REPLY, 0, &rc, &mh);
290         if (error)
291                 return;
292
293         error = dlm_dir_lookup(ls, nodeid, rc_in->rc_buf, len, &ret_nodeid);
294         if (error)
295                 ret_nodeid = error;
296         rc->rc_result = ret_nodeid;
297         rc->rc_id = rc_in->rc_id;
298         rc->rc_seq_reply = rc_in->rc_seq;
299
300         send_rcom(ls, mh, rc);
301 }
302
303 static void receive_rcom_lookup_reply(struct dlm_ls *ls, struct dlm_rcom *rc_in)
304 {
305         dlm_recover_master_reply(ls, rc_in);
306 }
307
308 static void pack_rcom_lock(struct dlm_rsb *r, struct dlm_lkb *lkb,
309                            struct rcom_lock *rl)
310 {
311         memset(rl, 0, sizeof(*rl));
312
313         rl->rl_ownpid = cpu_to_le32(lkb->lkb_ownpid);
314         rl->rl_lkid = cpu_to_le32(lkb->lkb_id);
315         rl->rl_exflags = cpu_to_le32(lkb->lkb_exflags);
316         rl->rl_flags = cpu_to_le32(lkb->lkb_flags);
317         rl->rl_lvbseq = cpu_to_le32(lkb->lkb_lvbseq);
318         rl->rl_rqmode = lkb->lkb_rqmode;
319         rl->rl_grmode = lkb->lkb_grmode;
320         rl->rl_status = lkb->lkb_status;
321         rl->rl_wait_type = cpu_to_le16(lkb->lkb_wait_type);
322
323         if (lkb->lkb_bastfn)
324                 rl->rl_asts |= AST_BAST;
325         if (lkb->lkb_astfn)
326                 rl->rl_asts |= AST_COMP;
327
328         rl->rl_namelen = cpu_to_le16(r->res_length);
329         memcpy(rl->rl_name, r->res_name, r->res_length);
330
331         /* FIXME: might we have an lvb without DLM_LKF_VALBLK set ?
332            If so, receive_rcom_lock_args() won't take this copy. */
333
334         if (lkb->lkb_lvbptr)
335                 memcpy(rl->rl_lvb, lkb->lkb_lvbptr, r->res_ls->ls_lvblen);
336 }
337
338 int dlm_send_rcom_lock(struct dlm_rsb *r, struct dlm_lkb *lkb)
339 {
340         struct dlm_ls *ls = r->res_ls;
341         struct dlm_rcom *rc;
342         struct dlm_mhandle *mh;
343         struct rcom_lock *rl;
344         int error, len = sizeof(struct rcom_lock);
345
346         if (lkb->lkb_lvbptr)
347                 len += ls->ls_lvblen;
348
349         error = create_rcom(ls, r->res_nodeid, DLM_RCOM_LOCK, len, &rc, &mh);
350         if (error)
351                 goto out;
352
353         rl = (struct rcom_lock *) rc->rc_buf;
354         pack_rcom_lock(r, lkb, rl);
355         rc->rc_id = (unsigned long) r;
356
357         send_rcom(ls, mh, rc);
358  out:
359         return error;
360 }
361
362 /* needs at least dlm_rcom + rcom_lock */
363 static void receive_rcom_lock(struct dlm_ls *ls, struct dlm_rcom *rc_in)
364 {
365         struct dlm_rcom *rc;
366         struct dlm_mhandle *mh;
367         int error, nodeid = rc_in->rc_header.h_nodeid;
368
369         dlm_recover_master_copy(ls, rc_in);
370
371         error = create_rcom(ls, nodeid, DLM_RCOM_LOCK_REPLY,
372                             sizeof(struct rcom_lock), &rc, &mh);
373         if (error)
374                 return;
375
376         /* We send back the same rcom_lock struct we received, but
377            dlm_recover_master_copy() has filled in rl_remid and rl_result */
378
379         memcpy(rc->rc_buf, rc_in->rc_buf, sizeof(struct rcom_lock));
380         rc->rc_id = rc_in->rc_id;
381         rc->rc_seq_reply = rc_in->rc_seq;
382
383         send_rcom(ls, mh, rc);
384 }
385
386 /* If the lockspace doesn't exist then still send a status message
387    back; it's possible that it just doesn't have its global_id yet. */
388
389 int dlm_send_ls_not_ready(int nodeid, struct dlm_rcom *rc_in)
390 {
391         struct dlm_rcom *rc;
392         struct rcom_config *rf;
393         struct dlm_mhandle *mh;
394         char *mb;
395         int mb_len = sizeof(struct dlm_rcom) + sizeof(struct rcom_config);
396
397         mh = dlm_lowcomms_get_buffer(nodeid, mb_len, GFP_NOFS, &mb);
398         if (!mh)
399                 return -ENOBUFS;
400         memset(mb, 0, mb_len);
401
402         rc = (struct dlm_rcom *) mb;
403
404         rc->rc_header.h_version = (DLM_HEADER_MAJOR | DLM_HEADER_MINOR);
405         rc->rc_header.h_lockspace = rc_in->rc_header.h_lockspace;
406         rc->rc_header.h_nodeid = dlm_our_nodeid();
407         rc->rc_header.h_length = mb_len;
408         rc->rc_header.h_cmd = DLM_RCOM;
409
410         rc->rc_type = DLM_RCOM_STATUS_REPLY;
411         rc->rc_id = rc_in->rc_id;
412         rc->rc_seq_reply = rc_in->rc_seq;
413         rc->rc_result = -ESRCH;
414
415         rf = (struct rcom_config *) rc->rc_buf;
416         rf->rf_lvblen = cpu_to_le32(~0U);
417
418         dlm_rcom_out(rc);
419         dlm_lowcomms_commit_buffer(mh);
420
421         return 0;
422 }
423
424 static int is_old_reply(struct dlm_ls *ls, struct dlm_rcom *rc)
425 {
426         uint64_t seq;
427         int rv = 0;
428
429         switch (rc->rc_type) {
430         case DLM_RCOM_STATUS_REPLY:
431         case DLM_RCOM_NAMES_REPLY:
432         case DLM_RCOM_LOOKUP_REPLY:
433         case DLM_RCOM_LOCK_REPLY:
434                 spin_lock(&ls->ls_recover_lock);
435                 seq = ls->ls_recover_seq;
436                 spin_unlock(&ls->ls_recover_lock);
437                 if (rc->rc_seq_reply != seq) {
438                         log_debug(ls, "ignoring old reply %x from %d "
439                                       "seq_reply %llx expect %llx",
440                                       rc->rc_type, rc->rc_header.h_nodeid,
441                                       (unsigned long long)rc->rc_seq_reply,
442                                       (unsigned long long)seq);
443                         rv = 1;
444                 }
445         }
446         return rv;
447 }
448
449 /* Called by dlm_recv; corresponds to dlm_receive_message() but special
450    recovery-only comms are sent through here. */
451
452 void dlm_receive_rcom(struct dlm_ls *ls, struct dlm_rcom *rc, int nodeid)
453 {
454         int lock_size = sizeof(struct dlm_rcom) + sizeof(struct rcom_lock);
455
456         if (dlm_recovery_stopped(ls) && (rc->rc_type != DLM_RCOM_STATUS)) {
457                 log_debug(ls, "ignoring recovery message %x from %d",
458                           rc->rc_type, nodeid);
459                 goto out;
460         }
461
462         if (is_old_reply(ls, rc))
463                 goto out;
464
465         switch (rc->rc_type) {
466         case DLM_RCOM_STATUS:
467                 receive_rcom_status(ls, rc);
468                 break;
469
470         case DLM_RCOM_NAMES:
471                 receive_rcom_names(ls, rc);
472                 break;
473
474         case DLM_RCOM_LOOKUP:
475                 receive_rcom_lookup(ls, rc);
476                 break;
477
478         case DLM_RCOM_LOCK:
479                 if (rc->rc_header.h_length < lock_size)
480                         goto Eshort;
481                 receive_rcom_lock(ls, rc);
482                 break;
483
484         case DLM_RCOM_STATUS_REPLY:
485                 receive_sync_reply(ls, rc);
486                 break;
487
488         case DLM_RCOM_NAMES_REPLY:
489                 receive_sync_reply(ls, rc);
490                 break;
491
492         case DLM_RCOM_LOOKUP_REPLY:
493                 receive_rcom_lookup_reply(ls, rc);
494                 break;
495
496         case DLM_RCOM_LOCK_REPLY:
497                 if (rc->rc_header.h_length < lock_size)
498                         goto Eshort;
499                 dlm_recover_process_copy(ls, rc);
500                 break;
501
502         default:
503                 log_error(ls, "receive_rcom bad type %d", rc->rc_type);
504         }
505 out:
506         return;
507 Eshort:
508         log_error(ls, "recovery message %x from %d is too short",
509                           rc->rc_type, nodeid);
510 }
511